Validation of pure rotational Raman temperature data from the Raman Lidar for Meteorological Observations (RALMO) at Payerne

Page created by Andrew Newton
 
CONTINUE READING
Validation of pure rotational Raman temperature data from the Raman Lidar for Meteorological Observations (RALMO) at Payerne
Atmos. Meas. Tech., 14, 1333–1353, 2021
https://doi.org/10.5194/amt-14-1333-2021
© Author(s) 2021. This work is distributed under
the Creative Commons Attribution 4.0 License.

Validation of pure rotational Raman temperature data
from the Raman Lidar for Meteorological
Observations (RALMO) at Payerne
Giovanni Martucci1 , Francisco Navas-Guzmán1,3 , Ludovic Renaud1 , Gonzague Romanens1 ,
S. Mahagammulla Gamage2 , Maxime Hervo1 , Pierre Jeanneta, , and Alexander Haefele1,2
1 Federal Office of Meteorology and Climatology, MeteoSwiss, Payerne, Switzerland
2 Department   of Physics and Astronomy, The University of Western Ontario, London, Canada
3 Andalusian Institute for Earth System Research, IISTA-CEAMA, University of Granada,

Junta de Andalucía, Granada 18006, Spain
a formerly at: Federal Office of Meteorology and Climatology, MeteoSwiss, Payerne, Switzerland
 retired

Correspondence: Giovanni Martucci (giovanni.martucci@meteoswiss.ch)

Received: 16 July 2020 – Discussion started: 10 September 2020
Revised: 8 December 2020 – Accepted: 11 December 2020 – Published: 22 February 2021

Abstract. The Raman Lidar for Meteorological Observa-             0.28, 0.02 ± 0.1 and 0.62 ± 0.03 K. The nighttime statis-
tions (RALMO) is operated at the MeteoSwiss station of            tics provide information for the entire troposphere and yield
Payerne (Switzerland) and provides, amongst other products,       1Tmax = 0.29 K, µ = 0.05 ± 0.34 K and σ = 0.66 ± 0.06 K.
continuous measurements of temperature since 2010. The            The small 1Tmax , µ and σ values obtained for both day-
temperature profiles are retrieved from the pure rotational       time and nighttime comparisons indicate the high stability
Raman (PRR) signals detected around the 355 nm Cabannes           of RALMO that has been calibrated only seven times over
line. The transmitter and receiver systems of RALMO are           18 months. The retrieval method can correct for the largest
described in detail, and the reception and acquisition units      sources of correlated and uncorrelated errors, e.g. signal
of the PRR channels are thoroughly characterized. The Fast-       noise, dead time of the acquisition system and solar back-
Com P7888 card used to acquire the PRR signal, the calcula-       ground. Especially the solar radiation (scattered into the field
tion of the dead time and the desaturation procedure are also     of view from the zenith angle 8) affects the quality of PRR
presented. The temperature profiles retrieved from RALMO          signals and represents a source of systematic error for the
PRR data during the period going from July 2017 to the end        retrieved temperature. An imperfect subtraction of the back-
of December 2018 have been validated against two refer-           ground from the daytime PRR profiles induces a bias of up
ence operational radiosounding systems (ORSs) co-located          to 2 K at all heights. An empirical correction f (8) ranging
with RALMO, i.e. the Meteolabor SRS-C50 and the Vaisala           from 0.99 to 1 has therefore been applied to the mean back-
RS41. The ORSs have also served to perform the calibra-           ground of the PRR signals to remove the bias. The correc-
tion of the RALMO temperature during the validation period.       tion function f (8) has been validated against the numerical
The maximum bias (1Tmax ), mean bias (µ) and mean stan-           weather prediction model COSMO (Consortium for Small-
dard deviation (σ ) of RALMO temperature Tral with respect        scale Modelling), suggesting that f (8) does not introduce
to the reference ORS, Tors , are used to characterize the accu-   any additional source of systematic or random error to Tral .
racy and precision of Tral along the troposphere. The daytime     A seasonality study has been performed to help with under-
statistics provide information essentially about the lower tro-   standing if the overall daytime and nighttime zero bias hides
posphere due to lower signal-to-noise ratio. The 1Tmax , µ        seasonal non-zero biases that cancel out when combined in
and σ of the differences 1T = Tral − Tors are, respectively,      the full dataset.

Published by Copernicus Publications on behalf of the European Geosciences Union.
Validation of pure rotational Raman temperature data from the Raman Lidar for Meteorological Observations (RALMO) at Payerne
1334                                            G. Martucci et al.: RALMO pure rotational Raman temperature validation

1   Introduction                                                   photon-count profiles are proportional to the atmospheric
                                                                   mass-density profile in an atmosphere that behaves like an
Continuous measurements of tropospheric temperature are            ideal gas and that is in hydrostatic equilibrium. The mass-
essential for numerous meteorological applications and in          density profile is used to determine the absolute temperature
particular for numerical weather predictions, for satellite cal-   profile (Hauchecorne et al., 1991; Alpers et al., 2004; Ar-
ibration and validation applications (Stiller et al., 2012; Wing   gall, 2007). The pure rotational Raman (PRR) method re-
et al., 2018), and for the understanding of climate change.        lies on the dependence of the rotational spectrum on atmo-
Co-located temperature and humidity measurements allow us          spheric temperature (Cooney, 1972; Vaughan et al., 1993;
to calculate the relative humidity, a parameter playing a key      Balin et al., 2004; Behrendt et al., 2004; Di Girolamo et al.,
role in several thermodynamic processes, such as the hygro-        2004; Achtert et al., 2013; Zuev et al., 2017). A combina-
scopic growth of condensation nuclei, fog and cloud forma-         tion of the Rayleigh and Raman methods is also possible
tion. When considering the thermodynamic processes occur-          and allows us to extend significantly the atmospheric region
ring within a stagnant air mass, a strong increase in relative     where the temperature is retrieved (Li et al., 2016; Gerding
humidity is often a precursor of fog, while the onset of super-    et al., 2008). The four methods have the common objective
saturation is linked to a consolidated radiation fog or a cloud    to produce a temperature profile as close as possible to the
forming at the top of a convective layer. Another important        true atmospheric status. In the attempt to do that, a reference
thermodynamic parameter is the convective available poten-         must be used to calibrate the lidar temperature and calcu-
tial energy (CAPE); CAPE is directly related to the temper-        late the related uncertainty. Trustworthy references can be
ature difference between two layers in the atmosphere. The         provided by co-located radiosondes, satellites or numerical
knowledge of temperature as a function of altitude allows          models. A co-located radiosounding system (RS) can act as
us to monitor the atmospheric thermodynamic stability and          reference to calibrate and monitor the stability of a lidar sys-
to diagnose and forecast the onset and intensity of a thun-        tem over long periods of time (Newsom et al., 2013). Our
derstorm. Despite its importance in all these processes, the       study presents a characterization of the RSs in use at Payerne
atmospheric temperature is still undersampled in the lower         and their validation with respect to the Vaisala RS92 certi-
troposphere where the traditional and well-established ob-         fied by the Global Climate Observing System (GCOS) Ref-
serving systems (e.g. radiosounding, Aircraft Meteorologi-         erence Upper-Air Network (GRUAN). Assimilation experi-
cal DAta Relay (AMDAR), MODE-S (Selective) EHS (en-                ments using validated Raman lidar temperature profiles have
hanced surveillance), satellites) do not provide continuous        been performed, among others, by Adam et al. (2016) and
measurements. A vertical profile of temperature in the tro-        Leuenberger et al. (2020). Both studies highlight the great
posphere can be measured efficiently by ground-based re-           potential of Raman lidar to improve numerical weather pre-
mote sensing instrumentation; unlike other technologies, re-       diction (NWP) models through data assimilation (DA).
mote sensing is best suited to operate continuously and to            In this study we characterize and validate Raman Lidar for
satisfy real-time data delivery requirements. Moreover, re-        Meteorological Observations (RALMO) temperature profiles
mote sensing instruments operating continuously for many           and demonstrate the high stability of the system. The paper
years ensure long time series of data, which are fundamental       is organized as follows. In Sect. 2 we establish the quality
for climatology studies.                                           of the reference radiosonde datasets. The lidar system is de-
   This study focuses on the measurement of the atmospheric        scribed in detail in Sect. 3 followed by a thorough uncer-
temperature done by a light detection and ranging (lidar)          tainty budget estimation in Sect. 4. In that section all con-
instrument. Best known methodologies to retrieve temper-           tributions to the total error are quantified and taken into ac-
ature profiles using a lidar can be split into four groups of      count (dead-time correction, background correction, photon-
techniques: the differential absorption lidar (DIAL), the high     counting error, calibration error). In Sects. 5 and 6 we present
spectral resolution lidar (HSRL), the Rayleigh technique and       the statistics of the comparison between lidar and radioson-
the Raman technique (Wulfmeyer et al., 2015, and refer-            des. In Sect. 5 we present the statistical analysis of the
ences therein). Measurements with DIAL are based on the            1T = Tral − Tors dataset, and we analyse the possible causes
dependency of the molecular absorption on the atmospheric          of µ and σ over the period July 2017–December 2018. An
temperature; namely, oxygen molecules with their constant          additional statistical study has been performed by splitting
mixing ratio in the dry atmosphere are used as targets by          the 1T dataset into seasons to investigate the effect of solar
DIAL to retrieve the temperature profile (Behrendt, 2005;          background and its correction function f (8) on the retrieved
Hua et al., 2005). The HSRL technique uses the Doppler             temperature profiles in terms of µ and σ (Sect. 6). The max-
frequency shifts produced when photons are scattered from          imum (1Tmax ) and mean bias (µ) of the difference (1T )
molecules in random thermal motion; the temperature depen-         of the lidar temperature profiles with respect to the tempo-
dence of the shape of the Cabannes line is used directly for       rally and spatially co-located radiosonde profiles represent
temperature measurements (Theopold and Bösenberg, 1993;            the systematic uncertainty of the lidar temperature. The stan-
Wulfmeyer and Bösenberg, 1998; Bösenberg, 1998). The               dard deviation of all differences 1T over the entire dataset
Rayleigh method is based on the assumption that measured           yields the random uncertainty (σ ) of the lidar temperature.

Atmos. Meas. Tech., 14, 1333–1353, 2021                                             https://doi.org/10.5194/amt-14-1333-2021
Validation of pure rotational Raman temperature data from the Raman Lidar for Meteorological Observations (RALMO) at Payerne
G. Martucci et al.: RALMO pure rotational Raman temperature validation                                                   1335

2   Validation of the reference radiosounding systems            0.15 ± 0.05 K. For the nighttime comparisons (23:00 UTC),
                                                                 the mean bias of RS41 over the region 0–14 km is −0.05 ±
In the framework of the operational radiosonde flight pro-       0.02 K with a mean standard deviation of 0.11 ± 0.06 K. The
gramme, the operational radiosonde is launched twice daily       statistics of SRS-C50 for daytime (11:00 UTC) show a mean
at Payerne at 11:00 and 23:00 UTC (in order to reach 100 hPa     bias over the region 0–14 km of −0.08 ± 0.02 K and a mean
by 00:00 and 12:00 UTC) and provides profiles of humid-          standard deviation of 0.19 ± 0.09 K. For the nighttime com-
ity (q), temperature (T ), pressure (P ) and wind (u). In ad-    parisons (23:00 UTC), the mean bias of SRS-C50 over the
dition to the operational programme, MeteoSwiss is part of       region 0–14 km is −0.01 ± 0.02 K with a mean standard de-
the Global Climate Observing System (GCOS) Reference             viation of 0.13 ± 0.04 K.
Upper-Air Network (GRUAN) since 2012 with the Vaisala               The comparisons with RS92 show that for both ORSs the
sonde RS92. In the framework of GRUAN, MeteoSwiss has            daytime differences undergo a larger variability along the
launched from the aerological station of Payerne more than       0–14 km vertical range compared to the nighttime statistics.
300 RS92 sondes between 2012 and 2019, contributing sig-         The main reason for the larger variability is that, during the
nificantly to its characterization (metadata, correction algo-   daytime flights, RS92 and the two ORSs undergo different
rithms and uncertainty calculation) and to its GRUAN certi-      exposures to the solar radiation, which causes a different re-
fication (Dirksen et al., 2014; Bodeker and Kremser, 2015).      sponse of the thermocouple sensors. The effect on the ther-
Before being part of GRUAN and since 2005, MeteoSwiss            mocouple becomes larger with altitude as the solar radiation
has used the RS92 sonde as working standard in the frame-        increases with height. All RSs are corrected by the manufac-
work of the quality assurance programme of the different         turer for the effects of solar radiation on the thermocouple
versions of the Meteolabor Swiss radiosonde (SRS). Differ-       sensors. However, different manufacturers use different radi-
ent versions of the SRS systems were operated at Payerne         ation corrections, which contributes to the statistical broad-
since 1990, starting from the analogue SRS-400 (from 1990        ening of the differences at all levels. The overall (11:00 and
to 2011) and developing to the digital sondes SRS-C34 and        23:00 UTC) performance of the two ORSs in terms of bias
C50. Starting from 2012, different versions of the SRS-C34       with respect to the reference RS92 is summarized in Fig. 3.
and SRS-C50 have been compared to RS92 in the framework          The distribution and mean value of the differences confirm
of GRUAN. In 2014, the Vaisala RS41 (Dirksen et al., 2020)       that in the first 15 km the two ORSs remain well below the
was added to the GRUAN programme where it performed              −0.1 K bias. The RS41 shows closer values to RS92 than
numerous multi-payload flights with RS92 and the SRS car-        SRS-C50 especially in the stratosphere. The better statistics
ried under the same balloon.                                     of RS41 should be interpreted also in light of the fact that
   During the studied period, two different operational ra-      RS92 and RS41 are both manufactured by Vaisala.
diosounding systems (ORSs) have been launched regularly
at 11:00 and 23:00 UTC: the SRS-C50 (February 2017–
March 2018) and the Vaisala RS41 (March–December 2018).          3   The Raman Lidar for Meteorological Observations –
Thanks to the multi-sensor flights performed with the SRS-           RALMO
C50, the Vaisala RS41 and the Vaisala RS92, the SRS-
C50 and RS41 have been validated by the GRUAN-certified          RALMO was designed and built by the École Polytech-
RS92. Figures 1 and 2 show the statistical biases of RS41 and    nique Fédérale de Lausanne (EPFL) in collaboration with
SRS-C50 with respect to the reference RS92 as a function of      MeteoSwiss. After its installation at the MeteoSwiss station
height for the day- and nighttime launches. The differences      of Payerne (46◦ 48.00 N, 6◦ 56.00 E; 491 m a.s.l.) in 2007 it
have been co-added into altitude boxes of 2 km, and the pro-     has provided profiles of q, T and aerosol backscatter (β)
files have been sampled every 30 s starting from 15 s after      in the troposphere and lower stratosphere almost uninter-
launch. The boxes in the plots have boundaries at the 25th       ruptedly since 2008 (Brocard et al., 2013; Dinoev et al.,
and 75th percentiles and are centred (black dot in each box)     2013). The T data during the 2008–2010 period are unex-
in the mean value bias.                                          ploited due to low quality of the analogue channel. RALMO
   The RS41 and SRS-C50 show an overall negative bias            has been designed to achieve a measurement precision bet-
during both day and night never exceeding −0.1 K along           ter than 10 % for q and 0.5 K for T with a 30 min inte-
the whole troposphere. Only in the mid-stratosphere, above       gration time and to reach at least 5 km during daytime and
30 km, does the daytime biases reach −0.5 K. In the frame-       7 km during nighttime in clear-sky conditions. RALMO uses
work of our study, the region of interest for the tempera-       high-energy emission, narrow field of view of the receiver
ture profiles measured by RALMO is the troposphere and,          and a narrowband detection to achieve the required daytime
more rarely, the UTLS (upper troposphere and lower strato-       performance. The data acquisition software has been devel-
sphere, ∼ 0–14 km). In this region, as the statistics show,      oped to ensure autonomous operation of the system and real-
the two ORSs perform very well. For the daytime compar-          time data availability. RALMO’s frequency-tripled Nd:YAG
isons (11:00 UTC), the mean bias of RS41 over the region 0–      laser emits 400 mJ per pulse at 30 Hz and at 355 nm. A
14 km is −0.05 ± 0.03 K with a mean standard deviation of        beam expander expands the beam’s diameter to 14 cm and re-

https://doi.org/10.5194/amt-14-1333-2021                                          Atmos. Meas. Tech., 14, 1333–1353, 2021
Validation of pure rotational Raman temperature data from the Raman Lidar for Meteorological Observations (RALMO) at Payerne
1336                                             G. Martucci et al.: RALMO pure rotational Raman temperature validation

Figure 1. Temperature deviation of RS41 with respect to RS92 calculated from the GRUAN multi-payload flights dataset during June
2015–December 2018. The boxes are centred in the mean bias and span the 25th–75th percentile range. Statistics are based on 58 flights at
11:00 UTC (a) and 59 flights at 23:00 UTC (b).

Figure 2. Temperature deviation of SRS-C50 with respect to RS92 calculated from the GRUAN multi-payload flights dataset during
February–December 2018. The boxes are centred in the mean bias and span the 25th–75th percentile range. Statistics are based on 25
flights at 11:00 UTC (a) and 26 flights at 23:00 UTC (b).

duces the beam divergence to 0.09±0.02 mrad. The returned                The Raman lidar equation, RLE, yields the intensity of the
signal is an envelope of the 355 nm elastic- and Raman-                PRR signal SPRR :
backscattered signals, i.e. PRR, water vapour, oxygen, ni-
trogen and Rayleigh. Next, the Raman lidar equation (RLE)                           C             2
                                                                       SPRR (z) =      O(z)n(z)0atm   (z)
for the PRR signal is presented along with the detailed de-                         z2
scription of how RALMO selects the high and low quantum-
                                                                                                                  !
                                                                                                        dσ i
                                                                                    X X                  
shifted wavelengths used in the RLE to retrieve the tempera-                                τ (Ji )ηi        (Ji ) + B.               (1)
                                                                                                        d 5
ture.                                                                             i=O ,N J
                                                                                      2   2   i

                                                                          The received SPRR signal measured over time t is a func-
3.1    Pure rotational Raman temperature                               tion of the altitude z; C is the lidar constant; O(z) is the geo-
                                                                       metrical overlap between the emitted laser and the receiver’s
Raman lidar measurements of the atmospheric temperature                field of view; n(z) is the number density of the air; 0atm (z)
rely on the interaction between the probing electromag-                is the atmospheric transmission; τ (Ji ) is the transmission of
netic signal at wavelength (λ) emitted by the lidar and the            the receiver for each PRR line Ji ; ηi is the volume mixing
                                                                                                           dσ i
molecules of O2 and N2 encountered along the probing                   ratio of nitrogen and oxygen; ( d     )5 (Ji ) is the differential
path. In addition to the Rayleigh light backscattered by the           Raman cross section for each PRR line Ji ; and B is the back-
aerosols and molecules at the same frequency as the inci-              ground of the measured signal. Air mainly contains oxygen
dent light, the O2 and N2 molecules return a frequency-                and nitrogen (≈ 99 %) whose ratio remains fairly constant in
shifted Raman signal back to the lidar’s receiver. The Raman-          the first 80 km of atmosphere, so ηi can be regarded as a con-
backscattered signal is shifted in frequency due to the rota-          stant in Eq. (1). The lidar constant C depends on the overall
tional and vibrational Raman effect. In this study only the            efficiency of the transceiver (transmitter and receiver) system
pure-rotational part of the spectrum around the Rayleigh fre-          including the photomultiplier tube (PMT) efficiency, on the
quency (Cabannes line) is detected by RALMO and analysed               area of the telescope and on the signal’s intensity. The full
(Fig. 4).                                                              expression of the differential Raman cross section for single

Atmos. Meas. Tech., 14, 1333–1353, 2021                                                   https://doi.org/10.5194/amt-14-1333-2021
Validation of pure rotational Raman temperature data from the Raman Lidar for Meteorological Observations (RALMO) at Payerne
G. Martucci et al.: RALMO pure rotational Raman temperature validation                                                                  1337

Figure 3. Overall temperature deviations at 11:00 and 23:00 UTC of RS41 (a) and SRS-C50 (b) with respect to the reference sonde RS92.
The interpretation of the graphics is the same as for the previous figures.

                                                                        polarized light. Nine-degree-tilted Semrock RazorEdge fil-
                                                                        ters (REFs) are installed just below the focal points of two
                                                                        of the four mirrors (left of Fig. 5). The REFs are long-
                                                                        wavelength pass filters and have a cut-off wavelength at
                                                                        364 nm (right of Fig. 5). The ro-vibrational Raman scattering
                                                                        from the atmospheric H2 O, O2 and N2 molecules is transmit-
                                                                        ted by the REF onto the optic fibres placed above the REF at
                                                                        the exact focal distance of the parabolic mirrors. The elastic
                                                                        (Rayleigh and Mie) and PRR scattering are reflected by the
                                                                        tilted REF onto 0.4 mm optic fibres and transmitted to the
                                                                        PRR polychromator.
                                                                            The two optic fibres transmitting the PRR and the elas-
                                                                        tic signals enter the temperature polychromator through the
                                                                        first fibre’s block shown in Fig. 6. The fibres are fixed into
                                                                        the fibre’s block, ensuring no or negligible temperature and
                                                                        mechanical-induced drifts of the fibre alignment with respect
Figure 4. Temperature dependence of the Stokes and anti-Stokes          to the other optical elements inside the polychromator (de-
pure rotational Raman spectrum of N2 calculated at 220 and 300 K.       tailed in Fig. 7). The outline of the two fibre block’s Carte-
The intensity of the spectral lines are in normalized relative units.   sian coordinates system in Fig. 6 shows the position of the in-
The frequency shift on the x axis refers to a laser’s wavelength of
                                                                        put, output and intermediate fibres (from stage 1 to stage 2)
355 nm. The total intensity of the high and low quantum-shifted
PRR channels is the summation of the single line intensities under-
                                                                        as a function of their abscissa–ordinate, x–y, positions. At
neath the bell-shaped black curves.                                     x = 21.5 mm, the input fibres, coming from the mirrors, are
                                                                        located at y = 20 mm and y = 24 mm; the output “elastic”
                                                                        fibres are located at y = 18 mm and y = 22 mm. The two
                                                                        output elastic signals are then transmitted through the fi-
lines of the PRR spectrum can be found in the reference book
                                                                        bres and combined together just before entering the PMT
chapter by Behrendt (2005).
                                                                        installed outside the polychromator’s box (PMT R12421P,
                                                                        Hamamatsu). The two input fibres transmit the PRR and the
3.2   Temperature polychromator of RALMO                                elastic signals onto an aspheric lens with focal length of
                                                                        300 mm and diameter of 150 mm. The two signals are then
The two-stage temperature polychromator, hereafter referred             transmitted through the lens onto a reflective holographic
to as PRR polychromator, represents the core of the signal              diffraction grating with groove density of 600 grooves mm−1
selection. The PRR polychromator separates several pure ro-             oriented at a diffraction angle of 48.15◦ with respect to the
tational Raman spectral lines and isolates elastic scattering           axis of the lenses in a Littrow configuration. The two input
consisting of Rayleigh and Mie lines (Cabannes line).                   signals (one from each mirror) are diffracted by the grat-
   The PRR signal from the O2 and N2 atmospheric                        ing polychromator and separated into high and low quan-
molecules is collected by four parabolic, high-efficiency re-           tum number lines from both Stokes and anti-Stokes parts
flecting mirrors each one with a diameter of 30 cm. The mir-            of the Raman-shifted spectrum. Two groups of four spectral
rors have a dielectric reflection coating with R > 99 % for the                                          Stokes , J anti-Stokes , J Stokes and
                                                                        lines are then diffracted, i.e. Jhigh      high            low
vibrational Raman wavelengths and R > 96 % for the elas-                  anti-Stokes .
tic and pure rotational Raman for both cross- and parallel-             Jlow

https://doi.org/10.5194/amt-14-1333-2021                                                   Atmos. Meas. Tech., 14, 1333–1353, 2021
Validation of pure rotational Raman temperature data from the Raman Lidar for Meteorological Observations (RALMO) at Payerne
1338                                                      G. Martucci et al.: RALMO pure rotational Raman temperature validation

Figure 5. (a) Telescopic mirrors, RazorEdge filter and optic fibres. (b) The REF cut-off frequency applied to the 355 nm Raman spectrum.

Table 1. Theoretical polychromator efficiencies (ξ ) for the nitrogen PRR lines at wavelengths λ.

                                                                              Nitrogen
                                                       Jlow                                           Jhigh
                                     anti-Stokes (n)
                                   Jlow                       λ           ξ         anti-Stokes (n)
                                                                                   Jhigh                      λ     ξ

                                   3                    354.2501       0.0325      10                  353.5532   0.2197
                                   4                    354.1503       0.2032      11                  353.4540   0.6577
                                   5                    354.0505       0.6270      12                  353.3549   0.9636
                                   6                    353.9509       0.9565      13                  353.2559   0.6915
                                   7                    353.8513       0.7217      14                  353.1570   0.2433
                                   8                    353.7519       0.2694      15                  353.0582   0.0420
                                     Stokes (n)
                                   Jlow                       λ           ξ         Stokes (n)
                                                                                   Jhigh                      λ     ξ

                                   3                    355.1511       0.0526      10                  355.8543   0.2577
                                   4                    355.2514       0.2709      11                  355.9548   0.6922
                                   5                    355.3518       0.7166      12                  356.0554   0.9500
                                   6                    355.4523       0.9735      13                  356.1560   0.6668
                                   7                    355.5527       0.6796      14                  356.2566   0.2395
                                   8                    355.6532       0.2439      15                  356.3572   0.0441

   The theoretical polychromator efficiencies ξ (ξ ∈ [0–1])                         J signals diffracted by the polychromator are then refocused
for the nitrogen and oxygen PRR lines Jhigh           Stokes , J anti-Stokes ,      by the aspheric lens onto the intermediate fibres positioned
                                                                 high
  Stokes and J anti-Stokes are shown in Tables 1 and 2, respec-
Jlow                                                                                at the y ordinates y = 18 mm and y = 22 mm and at the x ab-
                low
tively. The low and high quantum-number signals, Jlow and                           scissae x = 18.93 mm, x = 20.075 mm, x = 22.925 mm and
Jhigh , Raman-backscattered by the nitrogen molecules are                           x = 24.07 mm.
diffracted by the polychromator most efficiently at lines with                         The eight Stokes and anti-Stokes J signals are transmitted
quantum number n = 6 (Jlow      anti-Stokes , λ = 353.97 nm, J Stokes ,             through the intermediate fibres into the second fibre’s block
                                                                      low
λ = 355.47 nm) and n = 12 (Jhigh          anti-Stokes , λ = 353.37 nm,              (right of Fig. 6) and subsequently transmitted along an opti-
  Stokes , λ = 356.07 nm). Similarly to the nitrogen, the PRR                       cal path almost identical to the one in stage 1. Unlike stage 1,
Jhigh
                                                                                    the eight J signals are recombined by the diffraction grating
signals backscattered by the oxygen molecules are diffracted
                                                                                    polychromator into two groups of total J signals (Jhigh and
by the polychromator most efficiently at lines with quan-
                          anti-Stokes , λ = 353.96 nm, J Stokes , λ =               Jlow ). The general outline of the two-stage PRR polychroma-
tum number n = 9 (Jlow                                         low
                                anti-Stokes , λ = 353.38 nm, J Stokes ,             tor is shown in Fig. 7.
355.48 nm) and n = 17 (Jhigh                                          high             The total J signals are focused by the aspheric lens onto
λ = 356.06 nm).
                                                                                    the output fibres positioned in the second fibre’s block (right
   Signals Jlow and Jhigh are sums of the respective Stokes
                                                                                    of Fig. 6) at the same x abscissa x = 21.5 mm and at the
and anti-Stokes lines for nitrogen and oxygen. The eight
                                                                                    y ordinates y = 24.5 mm (Jhigh ), y = 21.5 mm (Jhigh ), y =

Atmos. Meas. Tech., 14, 1333–1353, 2021                                                                  https://doi.org/10.5194/amt-14-1333-2021
Validation of pure rotational Raman temperature data from the Raman Lidar for Meteorological Observations (RALMO) at Payerne
G. Martucci et al.: RALMO pure rotational Raman temperature validation                                                              1339

Table 2. Theoretical polychromator efficiencies (ξ ) for the oxygen PRR lines at wavelengths λ.

                                                                    Oxygen
                                                  Jlow                                     Jhigh
                                anti-Stokes (n)
                              Jlow                       λ      ξ        anti-Stokes (n)
                                                                        Jhigh                      λ     ξ

                              5                    354.2305   0.0493    15                  353.5124   0.3759
                              7                    354.0864   0.4535    17                  353.3696   0.9524
                              9                    353.9425   0.9598    19                  353.2272   0.5496
                              11                   353.7989   0.4686    21                  353.0851   0.0727
                                Stokes (n)
                              Jlow                       λ      ξ        Stokes (n)
                                                                        Jhigh                      λ     ξ

                              5                    355.1708   0.0765    15                  355.8956   0.4195
                              7                    355.3157   0.5454    17                  356.0404   0.9457
                              9                    355.4607   0.9690    19                  356.1852   0.5308
                              11                   355.6057   0.4301    21                  356.3298   0.0747

Figure 6. The outline of the two fibres’ blocks shows the positions of the input, output and intermediate (from stage 1 to stage 2) fibres
as a function of the abscissa–ordinate, x–y Cartesian coordinates. (a) Two optic fibres transmitting the PRR and the elastic signals enter
the temperature polychromator through the first fibre’s block. (b) The eight Stokes and anti-Stokes J signals are transmitted through the
intermediate fibres into the second fibre’s block. The eight J signals are recombined by the diffraction grating polychromator into two
groups of total J signals (Jhigh and Jlow ) and sent to the output fibres. Holes shown outside the fibre’s block are not fibre holders.

19 mm (Jlow ) and y = 16 mm (Jlow ). The output fibres trans-            3.3   PRR channel acquisition system
mit the four Jhigh and Jlow signals from the two mirrors to
two separate PMT boxes installed outside the polychroma-
                                                                         The acquisition of RALMO’s PRR channels were migrated
tor’s unit. Inside each PMT box, two J signals are combined
                                                                         in August 2015 from the Licel acquisition system to the
by an imaging system made by two lenses focusing onto a
                                                                         FAST ComTec P7888 (FastCom) system. The P7888 model
common spot. This last recombined signal is then divided by
                                                                         is one of the fastest commercially available multiple-event
a beam splitter into two signals, one at 10 % and the other at
                                                                         time digitizers with four inputs (one for each PRR chan-
90 % of the intensity, which are focused onto two indepen-
                                                                         nel) with very short acquisition system dead time and conse-
dent PMTs. A total of four signals are then obtained at the
                                 10 % , J 90 % , J 10 % and J 90 % .     quently minimum saturation effects of the photon-counting
end of the receiver chain, i.e. Jhigh    high     low        low         channels. Compared to the Licel acquisition system, Fast-
                                                                         Com acquires the PRR channels solely in photon-counting
                                                                         mode, with higher range resolution and with about twice

https://doi.org/10.5194/amt-14-1333-2021                                                      Atmos. Meas. Tech., 14, 1333–1353, 2021
Validation of pure rotational Raman temperature data from the Raman Lidar for Meteorological Observations (RALMO) at Payerne
1340                                              G. Martucci et al.: RALMO pure rotational Raman temperature validation

                                                                      Table 3. Spatial and temporal resolution of SPRR signals and Tral .

                                                                               Level                    δt     δzmin      δzmax
                                                                                                  (minute)    (metre)    (metre)
                                                                               Raw SPRR                  1        2.4         2.4
                                                                               Corrected SPRR            1        2.4         2.4
                                                                               Tral                     30        400          30

                                                                      an upper altitude cut-off is set at the altitude where the er-
                                                                      ror exceeds 0.75 K; in the presence of clouds, the upper limit
                                                                      is set by the cloud base detected by a co-located ceilometer.
Figure 7. Optics block diagram of PRR polychromator.                  Very often, the clear-sky cut-off altitude corresponds to an
                                                                      altitude between 5 and 7 km during daytime measurements.

shorter dead time, τ . The FastCom acquisition system ac-             4.1   Correction of SPRR
quires two low-transmission channels (Jhigh     10 % , J 10 % ) and
                                                        low
                                    90 %
two high-transmission channels (Jhigh        90 %
                                         , Jlow   ). Most photon-     The PRR signals are corrected for the systematic under-
counting acquisition systems are limited in performance by            estimation of the true photon-counting signal (dead time)
the dead time τ , i.e. the minimum amount of time in which            and for the offsets (instrumental and solar). The first cor-
two input signals may be resolved as separate events. When-           rection is then for the acquisition system’s dead time τ .
                                                                      The low-transmission channels Jhigh     10 % and J 10 % do not be-
ever two consecutive photons impinge on the detector with                                                               low
separation time t < τ , the system counts only one event. Cer-        come saturated and are used as reference channels to identify
                                                                                                                                90 %
tain types of acquisition systems can be corrected for the un-        the saturation of the high-transmission channels Jhigh         and
                                                                        90 %
derestimation induced by τ ; the correction of the PRR signals        Jlow   . Assuming that the PMTs and the associated electron-
measured by the FastCom system is presented in the Sect. 4.           ics obey the non-paralyzable assumption (Whiteman et al.,
                                                                      1992), we have studied the departure from the constant ratio
                                                                      J 10 % /J 90 % as a function of τ . We have applied a method
4   Retrieval of Tral and calculation of the uncertainty              based on the non-paralyzable condition (Newsom et al.,
                                                                      2009) to a year of data and have calculated τ for all the cases
The high- and low-frequency-shifted SPRR signals have the             when the saturation clearly affected the high-transmission
expression given in Eq. (1). In order to use them to retrieve         channels J 90 % .
the temperature profile, Eq. (1) shall be corrected for the              As soon as the saturation has its onset, the ratio
dead time and the background. Once the signals are cor-               J 10 % /J 90 % ceases to be constant and the saturated J 90 %
rected, their ratio is used to retrieve the temperature from          yields smaller count rates than the true ones. When the J 90 %
Eq. (3) scaled by two coefficients A and B. The atmospheric           is desaturated using the correct τ , it gives the τ -corrected sig-
temperature is then obtained from the calibration of Eq. (3)          nal Jdesat (τ ) = J 90 % /(1 − τ J 90 % ). One-thousand linear fits
with respect to Tors and the determination of A and B. The            J 10 % = f (Jdesat (τi )) are performed with τi varying within
calibrated temperature is then provided along with its uncer-         the interval τi ∈ [0–10] ns at steps of 0.01 ns. The linear fits
tainty. Table 3 summarizes the vertical and temporal resolu-          J 10 % = f (Jdesat (τi )) are performed over a temporal inter-
tion of the SPRR signal at different stages of the data pro-          val of 30 min and a vertical range defined by the count rates
cessing. The vertical resolution of Tral is not constant with         within the maximum range [0.5–50] MHz: Cmin and Cmax ,
altitude and depends on the calculated total random uncer-            respectively. For each linear fit, we calculate the value e(τi )
tainty in Eq. (6). A Savitzky–Golay digital filter with poly-         that provides the distance in the count-rate domain between
nomial degree K = 1 is applied to the Tral profiles to degrade        J 10 % and f (Jdesat (τi )) as a function of τi . The minimiza-
the sampling resolution and reduce the sampling noise. The            tion of e(τ ) with respect to τi determines the value of the ac-
adopted procedure and the definition of the obtained verti-           quisition system’s dead time τi = τmin ∈ [0–10] ns for each
cal resolution are compliant with the NDACC (Network for              channel. The obtained value τmin is used to desaturate J 90 %
the Detection of Atmospheric Composition Change) recom-               and to re-establish the constant ratio J 10 % /f (Jdesat (τmin )),
mendations detailed in the work by Leblanc et al. (2016).             which equals a constant. Figure 8 shows an example cal-
The initial and highest resolution is δzmax = 30 m, which is          culation of τmin for the high-transmission channel of Jlow .
degraded to a minimum δzmin = 400 m corresponding to the              The curve function e(τ ) in the figure’s left panel has a mini-
regions where the error is large (normally, the upper tropo-          mum at τmin = 3 ns. On the right panel, the uncorrected and
sphere and lower stratosphere). For clear-sky measurement,            the τ -corrected relations are shown. The uncorrected relation

Atmos. Meas. Tech., 14, 1333–1353, 2021                                                 https://doi.org/10.5194/amt-14-1333-2021
Validation of pure rotational Raman temperature data from the Raman Lidar for Meteorological Observations (RALMO) at Payerne
G. Martucci et al.: RALMO pure rotational Raman temperature validation                                                                 1341

Figure 8. Dead-time calculation and correction of the high-transmission channel J 90 % . Panel (a) shows the minimization of the distance
vector e(τi ) in terms of τi yielding the dead time τ = 2.99 ns. Panel (b) shows the saturated J 10 % = f (Jdesat (τ = 0)) (solid black), the
desaturated linear relation J 10 % = f (Jdesat (τmin )) (dashed green) and the count-rate domain [Cmin , Cmax ] (dashed red).

J 10 % = f (Jdesat (τ = 0)) (solid black) departs from the lin-          f (8) ∈ [99 %–100 %], reducing B by the maximum amount
                                                                                                                            year
ear relation J 10 % = f (Jdesat (τmin )) (dashed green) as soon          of 1 % (f (8) = 99 %) at noon on 21 June when 8 = 8min .
as the count rates exceed the lower bound Cmin (dashed red).
By applying this method to a year of data and collecting more                                  cos 8
than 100 cases, we have determined the mean dead times                   f (8) = 1 − 0.01 ·            δ8 ; δ8 ≡ 1 for 0 ≤ 8 < π/2,
                                                                                              cos 8min
τj h = 1.4 ns and τj l = 3 ns for Jhigh and Jlow , respectively.
                                                                         δ8 ≡ 0 for π/2 ≤ 8 < 2π.                                        (2)
The desaturated Jhigh and Jlow are further corrected for the
background and the procedure is described hereafter.                        If uncorrected, the retrieved daytime Tral suffers a bias
   The electronic and solar background must be subtracted                at all heights with respect to Tors . The bias is largest when
from SPRR before retrieving Tral . While the electronic back-                    day
                                                                         8 = 8min . The correction f (8) is applied only to the back-
ground is stable and does not undergo daily or seasonal cy-
                                                                         ground of Jhigh . The intensity of Jhigh is generally lower
cles, the solar background changes in intensity with the po-
                                                                         than Jlow at all atmospheric temperatures (see Sect. 4) and
sition of the sun 8 (the angle between the zenith and the
                                                                         so is its signal-to-noise ratio (SNR). Even a small error of
centre of the sun’s disc). We have found that subtracting the
                                                                         ≈ 1 % when subtracting B from Jhigh has a major impact on
mean value of the far-range signal (z ∈ [50–60] km) from
                                                                         its SNR; f (8) corrects the imperfect subtraction of B from
SPRR (subtraction of term B from Eq. 1) causes a system-
                                                                         Jhigh and minimizes the daytime bias of Tral with respect to
atic negative bias with respect to Tors of about 1 K at all alti-
                                                                         Tors almost perfectly.
tudes z during daytime. A relative change of 1 % in the ratio
Jlow /Jhigh due to an imperfect background subtraction can               4.2   Estimation of total random uncertainty
lead to a variation of up to 2 K in the retrieved temperature
Tral . Because the solar background (SB) dominates the to-               Once Bcorr is subtracted from the τ -corrected SPRR , the de-
tal background of SPRR , we focus on the correction of the               viation of Tral from Tors depends only on how precisely
background B only as a function of the position of the sun.              Eq. (3), derived from the RLE, represents the true atmo-
We have developed an empirical correction function f (8)                 spheric temperature at the altitude z and time t. The ran-
applied to the background prior to subtraction from SPRR .               dom uncertainty does not account for the error induced by
The function f (8) is applied to the background B and pro-               the saturation and the background, which are considered
vides the corrected background Bcorr = f (8) · B. Through                purely systematic. The high-frequency-shifted (Jhigh ) and
the year’s cycle, B is reduced by a maximum amount of                    low-frequency-shifted (Jlow ) signals in the Stokes and anti-
1 % via the action of f (8). As Eq. (2) shows, f (8) reaches             stokes Q branches depend on the temperature of the probed
                              day
daily minima when 8 = 8min (noon) and returns to 100 %                   atmospheric volume (Fig. 4). The ratio of the SPRR intensi-
               ◦
when 8 ≥ 90 (after sunset and before sunrise). During the                ties Q(z) = Jlow (z)/Jhigh (z) is a function of the atmospheric
daily and annual cycle, f (8) then oscillates within the range           temperature T at the distance z. Based on the calculations
                                                                         shown by Behrendt (2005) and for systems that can detect in-
                                                                         dependent J lines in each channel, the relationship between
                                                                         T and Q would take the form of Eq. (3) with an equals (=)

https://doi.org/10.5194/amt-14-1333-2021                                                      Atmos. Meas. Tech., 14, 1333–1353, 2021
Validation of pure rotational Raman temperature data from the Raman Lidar for Meteorological Observations (RALMO) at Payerne
1342                                                                    G. Martucci et al.: RALMO pure rotational Raman temperature validation

sign. The approximately equal to sign (≈) in Eq. (3) indi-                                     curring drift of the detection system’s sensitivity and/or effi-
cates that the detection system detects more than one J line                                   ciency. Calibrations of RALMO are performed using Eq. (3)
and thus brings an inherent error. The calibration coefficients                                in clear-sky conditions during nighttime to remove the effect
A and B are a priori undetermined and can be determined by                                     of solar background and have a larger vertical portion of Tral
calibration of Tral with respect to Tors .                                                     available for calibration (the daytime profiles have normally
                                                                                               a lower cut-off altitude). Figure 9 shows a case of RALMO
              A                                                                                calibration; the green-shaded area represents ±2UT (k = 2).
Tral ≈              .                                                                    (3)      The calibrated Tral results from the integration of 30 τ -
           B + ln Q
                                                                                               and B-corrected SPRR profiles (δt = 30 min) into Eq. (3).
   The coefficients A and B are determined by calibrating                                      At a given atmospheric altitude z during the time interval
Tral with respect to Tors . The coefficient A has units in kelvin                              δt, UT (z)|δt is made of the single contributions Usig (z, t)
as Eq. (3) is not normalized for the standard atmospheric                                      and Ufit (z, t). Usig (z, t) can be regarded as independent with
temperature (Behrendt and Reichardt, 2000). The linear re-                                     respect to time; on the other hand, the errors Ufit (z, t) de-
lation y = A/(x + B) is used, where x is the ratio Q, and                                      pend on the atmospheric processes occurring within the layer
y is the reference temperature Tors . The mean error on Tors                                   [z−δz/2, z+δz/2] during the time interval [t −δt/2, t +δt/2]
for both RS41 and SRS-C50 is ≈ −0.04 ± 0.15 K at 11:00                                         and are, a priori, not statistically independent. By assum-
and 23:00 UTC between 0 and 14 km (Sect. 2). Due to the                                        ing that all errors in Eq. (6) are statistically independent,
very small error on Tors , we can calculate the uncertainty Ufit                               we assume that the off-diagonal elements of the variance–
of the fitting model only in terms of the fitting parameters’                                  covariance matrix are all zero. By doing so, UT (z)|δt could
errors σA and σB (Eq. 4). As it will be shown in the next                                      be underestimated by an amount equal to the non-zero co-
section, the covariance σAB of A and B is very close to zero                                   variance terms, including the covariance σAB . A method to
(< 10−3 ); thus, σA and σB can be treated as statistically inde-                               assess the exhaustiveness of the theoretical error UT is to
pendent and used to calculate Ufit from the first-order Taylor                                 calculate how many points in the vector 1T = Tral − Tors fall
series of propagation of fitting parameter uncertainties.                                      within the interval [−kσ, +kσ ] and check if they are com-
                                                                                               patible with the Gaussian probability levels 68.3 %, 95.5 %
                  "                   #1                                                       and 99.7 % for k = 1, 2, and 3, respectively. As it is shown
                             A 2σ 2    2
           1                                                                                   in the right panel of Fig. 9, almost all points along the vector
Ufit =              σA2 +        B
                                         .                                               (4)
       (B + ln Q)         (B + ln Q)2                                                          Tral − Tors fall within the interval [−2σ, +2σ ], i.e. the 98.1 %
                                                                                               envelope. For k = 1, the percentage falls slightly below the
   Ufit is not the only error source; a second contribution to                                 expected level for a normal distribution with only 61.2 % of
the total uncertainty comes from the fact that SPRR is ac-                                     the points within [−σ, +σ ]. Between July 2017 and Decem-
quired by a photon-counting system and is affected by the                                      ber 2018, a total of seven calibrations have been performed
measurement’s noise that can be calculated using standard                                      (three SRS-C50 and four Vaisala RS41). The mean percent-
Poisson statistics. The error σJ for Poisson-distributed√data                                  age of points over all performed calibrations is 65.1 % for
is equal to thepsquare root of the SPRR signals, σJlow = Jlow                                  k = 1; 97.9.1 % for k = 2; and 99.98 % for k = 3. These val-
and σJhigh = Jhigh . In Eq. (5), coefficients A and B can be                                   ues seem to confirm an overall exhaustiveness of UT with a
regarded as independent from the noise on SPRR , as the con-                                   slight underestimation of 3.2 % at k = 1. The list of calibra-
tribution of it is already included in σA and σB in Eq. (4).                                   tions is shown in Table 4. For each calibration, the table lists
                                                                                               the date and end time of the calibration, the calibration coeffi-
                                    1                                                        cients A and B used in the fitting model Eq. (3), the errors σA
                  A         1    1 2
Usig =                         +        .                                                (5)   and σB , the covariance σAB , and the ORS used to calibrate
              (B + ln Q)2 Jhigh Jlow                                                           Tral . As to further support the assumption of zero covariance
   The total uncertainty of the calibrated Tral is a type-B un-                                of the coefficients A and B, all covariances σAB in the table
certainty (JCGM, 2008) and is the sum of the independent                                       are smaller than 1.71 × 10−3 . Between two consecutive cali-
error contributions Ufit and Usig .                                                            brations performed at times ti and ti+1 , the coefficients A(ti )
                                                                                               and B(ti ) are used to calibrate all profiles Tral during the time
          q                                                                                    interval t ∈ [ti , ti+1 ].
UT =             2 + U2
               Ufit   sig
          s                                                                      
                  1                                              1           1
      =                    2
                                   σA2 (B + ln Q)2 + A2 σB2 +           +            .   (6)   5   Validation of PRR temperature
              (B + ln Q)                                        Jhigh       Jlow

4.3       Calibration of SPRR                                                                  More than 450 profiles Tral (245 nighttime, 215 daytime)
                                                                                               have been compared to Tors and assessed separately for day-
For a very stable system like RALMO, calibrations can be                                       time and nighttime based on the bias and standard deviation
performed once every few months to compensate for any oc-                                      (σ ) of the differences 1T = Tral − Tors over the period 1 July

Atmos. Meas. Tech., 14, 1333–1353, 2021                                                                          https://doi.org/10.5194/amt-14-1333-2021
G. Martucci et al.: RALMO pure rotational Raman temperature validation                                                            1343

Figure 9. RALMO temperature profile calibrated by the ORS RS41. Panel (a) shows TRS41 (solid black) and Tral (solid red) with the
confidence interval (green shading) corresponding to kUT (k = 2). Panel (b) shows the differences 1Tral−RS41 within the UT k-boundaries
for k = 1 (dashed blue) and k = 2 (dashed red).

Table 4. Calibrations of RALMO T profiles by the working standard.

                      YYYY-MM-DD HH:MM             A [K]      B      σA [K]     σB          σAB        ORS
                      2017-07-06 23:30            372.97    0.42     0.7275   0.0027    0.78 × 10−3    SRS-C50
                      2017-08-24 23:30            375.63    0.43     1.1184   0.0041    1.71 × 10−3    SRS-C50
                      2017-10-16 23:30            376.44    0.42     0.5987   0.0022    0.93 × 10−3    SRS-C50
                      2018-04-21 23:30            372.94    0.41     0.6980   0.0026    0.76 × 10−3    RS41
                      2018-05-11 23:30            373.14    0.41     0.9510   0.0036    1.35 × 10−3    RS41
                      2018-07-07 23:30            374.42    0.41     0.9877   0.0037    1.46 × 10−3    RS41
                      2018-09-11 23:30            374.78    0.41     0.9276   0.0034    1.30 × 10−3    RS41

2017–31 December 2018. Two criteria to select the cases for           1T exceeding 5 K. If more than 33 % of the elements along
the dataset have been used:                                           1T exceed the threshold, the whole of Tral is rejected and
                                                                      not included in the statistics. For any value below the thresh-
  1. Only cases with no precipitation and no low clouds or
                                                                      old, the outliers are removed from Tral . This is justified by the
     fog are retained.
                                                                      fact that exceedances counting more than 33 % are caused by
  2. Only cases with 1T < 5 K are retained.                           temporary misalignment of the transceiver unit. On the other
                                                                      hand, exceedances well below the threshold can always oc-
   Criterion 1 is performed by setting a threshold for mini-          cur (especially in the higher part of the profile) due to low
mum cloud base, hb , at 1000 m (a.g.l.), for any values of hb <       SNR or unfiltered clouds. In Table 5 we present a summary
1000 m that Tral is not retrieved. Whenever hb > 1000 m, the          of the statistical parameters characterizing the daytime and
cut-off altitude will correspond to hb . Indeed, above hb , the       nighttime differences 1T that will be discussed in detail in
SNR drops abruptly and UT (z)  1 K. For this reason, espe-           the following sections. The dataset 1T is described in terms
cially during winter, when long-lasting stratus clouds occur          of maximum mean bias 1Tmax , average mean bias µ, stan-
in the altitude range 1–4 km, the nighttime and daytime Tral          dard deviation σ and maximum availability Nmax of the dif-
profiles are limited in range to the altitude hb (or are not cal-     ferences 1T along the atmospheric range.
culated if hb < 1000 m). Criterion 2 is performed by setting a
threshold at 33 % of the number of elements along the profile

https://doi.org/10.5194/amt-14-1333-2021                                                Atmos. Meas. Tech., 14, 1333–1353, 2021
1344                                                G. Martucci et al.: RALMO pure rotational Raman temperature validation

Table 5. Summary of the statistical parameters of the day and night   0.1 K, 0.62 ± 0.03 K and 212, respectively. The data avail-
1T dataset.                                                           ability goes rapidly to zero above 5 km.
                                                                         The daytime Tral profiles have been corrected for the solar
  1Tmax     µ               σ                Nmax     Range           background by f (8), which proves to be very efficient in re-
  Night                                                               moving the noon bias with respect to Tors . To ensure that the
                                                                      correction f (8) does not introduce any additional bias dur-
  0.24 K    0.05 ± 0.34 K   0.66 ± 0.06 K     244     0.5–10 km
                                                                      ing the daily cycle, we have compared Tral with the temper-
  Day                                                                 ature calculated by the COSMO model. More than 3 months
  0.25 K    0.02 ± 0.1 K    0.62 ± 0.03 K     212     0.5–6 km        of clear-sky 24 h Tral − Tcos differences have been collected,
                                                                      yielding a mean daily cycle at the level 1.4–1.7 km a.s.l. The
                                                                      comparison in Fig. 12 shows that RALMO does not suffer
                                                                      any systematic daily 8-dependent bias.
   Besides a global validation, we also present and discuss
the seasonal statistics in order to better characterize the sys-
tem performance.
                                                                      6     Seasonality study
5.1    Nighttime temperature statistics
                                                                      In order to study the seasonal effects on µ and σ , the 1T
The nighttime 1Tmax , µ, σ and Nmax of 1T are the met-                dataset was divided into seasons. The four seasons are de-
rics to assess the accuracy and precision of Tral with re-            fined as follows: summer from 1 June to 30 August, autumn
spect to Tors . Figure 10 shows that 1Tmax = 0.24 K, µ =              from 1 September to 30 November, winter from 1 December
0.05 ± 0.34 K, σ = 0.66 ± 0.06 K and Nmax = 244 over the              to 15 March, and spring from 16 March to 31 May. Because
tropospheric region [0.5–10] km.                                      of the less favourable conditions in winter due to precipita-
   In addition to the uncertainty assessment performed in             tion and low clouds, only few temperature profiles are avail-
Sect. 4.3, the exhaustiveness of the theoretical total uncer-         able during this period. Additionally, during winter 2018,
tainty UT can be further assessed by comparing UT with σ .            from January until mid-March, RALMO measurements have
The mean value of UT along the troposphere and over the               been stopped for about 80 % of the time due to maintenance
seven nighttime calibrations is U T = 0.64 K; the mean night-         work. The results are summarized in Tables 6 and 7 in terms
time standard deviation averaged over the tropospheric col-           of µ, σ and maximum availability Nmax over the lower tro-
umn in Fig. 10 is σ = 0.66 K. The two 1−k uncertainties are           pospheric range 0–6 km for daytime and over 0–10 km for
then fully compatible.                                                nighttime. With the only exception of winter, the other sea-
   The nighttime 1T data are characterized by values of µ             sons have enough profiles to perform a statistical analysis
and σ smaller than 1 K, with minimum values in the lower              and to draw quantitative conclusions about the contribution
troposphere from [0–5] km. It is indeed in the lower tropo-           of each season to the overall values µ and σ .
sphere, where σ is ≈ 0.6 K, i.e. 0.1 K larger than the 0.5 K
requirement for data assimilation into the numerical weather          6.1    Seasonal daytime temperature statistics
prediction COSMO forecasting system (Fuhrer et al., 2018;
Klasa et al., 2018, 2019). In order to achieve a successful           The seasonal daytime µ and σ profiles are analysed to un-
assimilation of Tral into COSMO, the overall impact of the            derstand if sources of systematic errors other than SB affect
assimilation shall correspond to an improvement of the fore-          the retrieved Tral . The seasonal profiles are shown in Figs. 13
casts without increasing the forecast uncertainties. Assimila-        and 14 and summarized in Table 6. Due to the less favourable
tion of high-SNR, well-calibrated Tral into numerical models          weather conditions and the maintenance work, the winter
leads to the improvement of the forecasts. In the study by            statistics count only eight profiles. The statistical characteri-
Leuenberger et al. (2020), the authors assimilate, amongst            zation of the winter dataset can then only be qualitative. Sum-
other data, temperature and humidity profiles from RALMO,             mer and spring are the seasons with the minimum values of
showing the beneficial impact on the precipitation forecast           8 at noon; during these two seasons, Tral is most affected by
over a wide geographical area.                                        SB. If uncorrected for f (8), the noon Tral suffers a negative
                                                                      mean bias of about 2 K at all heights (not shown here). Sum-
5.2    Daytime temperature statistics                                 mer counts more than twice the number of cases in the spring
                                                                      dataset; nevertheless, for both seasons the values of µ are
During daytime, the retrieved temperature profiles are lim-           compatible with a zero bias within their uncertainties (both
ited in range to about 6 km. The left and right panels of             σ = 0.64 K). Despite the less favourable weather conditions
Fig. 11 show the bias 1T and the standard deviation σ , re-           compared to spring and summer, autumn is the season with
spectively. The 1Tmax , µ, σ and Nmax of the daytime 1T               most cases, and this is because there are two autumn seasons
over the lower troposphere (0.5–6 km) are 0.25 K, 0.02 ±              in the dataset. Like spring and summer, autumn also has µ

Atmos. Meas. Tech., 14, 1333–1353, 2021                                                https://doi.org/10.5194/amt-14-1333-2021
G. Martucci et al.: RALMO pure rotational Raman temperature validation                                                                  1345

Figure 10. Nighttime bias and standard deviation (SD) of 1T over the period July 2017–December 2018. In (a), on each box of 200 m vertical
span, the central mark indicates the median, and the left and right edges of the box indicate the 25th and 75th percentiles, respectively. The
whiskers extend to the most extreme data points not considered outliers, and the outliers are plotted individually and shown by the “+”
symbol. The thick red curve in the right panel shows the vertical profile of SD calculated over the altitude-decreasing number of 1T points
(thin green).

Table 6. Seasonal mean bias µ, standard deviation σ and maximum          mer and autumn σ profiles in the upper-left and upper-right
availability Nmax at 12:00 UTC.                                          panels undergo a decoupling between the lower and upper
                                                                         part of the profile with inversion point slightly higher in au-
               Season       µ [K]     σ [K]       Nmax                   tumn. An increase in σ with height is something expected
                                0.5–6 km a.s.l.                          and can be explained with the decreased data availability and
                                                                         the decreased SNR due to the large distance from the lidar’s
               Summer       +0.03     0.64          74                   telescope. However, the abrupt increase in σ at ≈ 3 km in
               Autumn      +0.005     0.61         102                   summer and at ≈ 4.5 km in autumn is more related to the at-
               Winter       +0.24     0.41           8
                                                                         mospheric dynamics than to the SNR. In summer, the transi-
               Spring       −0.03     0.64          31
                                                                         tion between the boundary layer and the free troposphere is a
                                                                         region of high variability in terms of temperature and humid-
                                                                         ity. The alternating cold downdraughts and warm updraughts
compatible with the zero bias within its uncertainty. Through            engendered by the overall fair weather conditions and the
the four seasons, the mean σ spans from 0.4 to 0.65 K.                   continuous development of thermals through the boundary
   Figure 13 suggests that Tral is not affected by any obvious           layer (Martucci et al., 2010) cause a large variability of Tral
systematic error (µ ' 0 ∈ [−σ, +σ ]) and no seasonal cycle               at ≈ 3 km, which translates into large σ values. In autumn,
appears in the statistics. From the perspective of the statisti-         the thermal activity at the top of the boundary layer is less
cal validity of the studied data, any subsample chosen ran-              pronounced than in summer; on the other hand, a tempera-
domly from the total Tral dataset can be described by the                ture inversion linked to the formation and dissipation of stra-
same µ and σ that characterizes 1T .                                     tus clouds above Payerne occurs at ≈ 4.5 km, causing larger
   Unlike Fig. 13, the σ profiles in Fig. 14 show different be-          discrepancies in the comparison with Tors .
haviours in summer–autumn and in winter–spring. The sum-

https://doi.org/10.5194/amt-14-1333-2021                                                     Atmos. Meas. Tech., 14, 1333–1353, 2021
1346                                               G. Martucci et al.: RALMO pure rotational Raman temperature validation

Figure 11. Daytime bias and standard deviation (SD) of 1T over the period July 2017–December 2018. In (a), on each box of 200 m vertical
span, the central mark indicates the median, and the left and right edges of the box indicate the 25th and 75th percentiles, respectively.
The whiskers extend to the most extreme data points not considered outliers, and the outliers are plotted individually and shown by the “+”
symbol. Panel (b) shows the vertical profile of standard deviation (thick red) calculated over the altitude-decreasing number of 1T points
(thin green).

Figure 12. Differences between RALMO and COSMO temperature at 1400–1700 m a.s.l. during September 2017–January 2018. The green
shading accounts for the k = 1 standard deviation of the differences. The dashed vertical lines show the sunrise and sunset time on 12 January
2018; the mean daily cycle shows that no artefacts are introduced by the application of f (8) during daylight.

6.2    Seasonal nighttime temperature statistics                         ration into seasons helps with understanding if the overall
                                                                         zero bias shown in Fig. 10 hides seasonal non-zero biases
                                                                         that cancel out when combined in the full dataset. Compared
At nighttime, f (8) has no impact on the temperature re-
                                                                         to the daytime cases (215), the availability of the nighttime
trieval, and the seasonal statistics can reveal sources of sys-
tematic error other than the SB causing |µ| > 0. The sepa-

Atmos. Meas. Tech., 14, 1333–1353, 2021                                                     https://doi.org/10.5194/amt-14-1333-2021
G. Martucci et al.: RALMO pure rotational Raman temperature validation                                                                  1347

Figure 13. Seasonal daytime bias of 1T over the period July 2017–December 2018. The boxplot characteristics are the same as in Figs. 10
and 11 but restricted over the seasonal periods. Based on the definition of seasons provided in the text, in clockwise order starting form the
upper-left corner, panel (a) shows the summer data, panel (b) shows the autumn data, panel (c) shows the winter data and panel (d) shows
the spring data.

dataset is higher (245), including the winter dataset, which             vious source of systematic error. The mean µ and σ in the
allows us to perform a statistical analysis for all seasons.             troposphere are summarized in Table 7.
   All seasonal µ values are compatible with the zero bias                  The nighttime atmosphere undergos different dynamics
along the troposphere within [−σ, +σ ]. Like the daytime                 with respect to daytime. The absence of solar radiation al-
seasonal statistics, the nighttime also does not reveal any ob-          most entirely removes the convection from the boundary
                                                                         layer and minimizes the variance of the temperature at the top

https://doi.org/10.5194/amt-14-1333-2021                                                     Atmos. Meas. Tech., 14, 1333–1353, 2021
You can also read